Bandwidth distribution (was: Re: AllowInvalidNodes entry, exit, ... ?)

[Sven Anderson <sven@xxxxxxxxxxx>]

Hi Mac,

Am 18.08.2008 um 16:43 schrieb macintoshzoom:

> Using "valid nodes" I have noticed too many times mu browsing is  
> going to the same exit nodes .... yes fast, but always the same tor  
> exit nodes "club".

this is not really a surprise if you look at the distribution of the  
bandwidth. I did some graphs for the bandwidth distribution of  
yesterday.

As you can see in [1] the distribution of bandwidth over the exit  
nodes follows a power-law (aka Pareto, Zipf, heavy/long tail, ...),  
like so many other distributions. In the double-logarithmic plot this  
is expressed in a linear relation. In this case the linearity starts  
between 20 and 30 kB/s. (The bandwidth of the exit nodes is  
exponentially binned which results in the equidistant data points.)

These power-law distributions have the well-known characteristic of  
many small values and very few big values, also referred to as 90/10  
or 80/20 rule. In plot [2] you can see the cumulative distribution  
function (CDF) over the ranked exit nodes. As you can see, the 30  
biggest exit nodes are holding 50% of the total tor exit bandwidth,  
and the 100 biggest hold 70%. While this is still quite moderate it  
shows how often you will see the top 30, even if the exit node  
selection would only be based on bandwidth. But the "Fast" and  
"Stable" flags of course increase this effect.

So there's no conspiracy, it's a natural law.

[1] http://sven.anderson.de/misc/en_bw_dist.pdf
[2] http://sven.anderson.de/misc/en_bw_cdf.pdf


Sven

--
http://sven.anderson.de    "Believe those who are seeking the truth.
tel:    +49-551-9969285     Doubt those who find it."
mobile: +49-179-4939223                                 (André Gide)